Combining Constraints and Consistency Techniques in Knowledge-Based Expert Systems

• Authors: Ilié Popescu
• Venue: ADVIS '00 Proceedings of the First International Conference on Advances in Information Systems
• Year: 2000

Knowledge-Based Expert Systems (KBES) have long been widely used to perform tasks that normally require human knowledge and intelligence. One important issue that has not been addressed satisfactorily in the existing KBESs is that they try to make posing queries simple by letting the users specify what they want to compute rather than how to compute it. In this paper, we show that the solutions computation process can be modeled with Constraint Satisfaction Problem (CSP) techniques, employing their simple representation schemes and consistency techniques. The motivation behind this is the desire to build up a computation model for reducing the vast amount of deductions required by a KBES when executed on a logic program system. A key idea is to represent the relations among the rules as constraints and to integrate the rule chaining with constraint solving. In this integration, the constraints are regarded as special facts at each node of the solutions graph, and the constraints propagation may cause firing of rules. In this way the model allows the solutions graph to grow progressively by enumerating the solutions of the system of constraints and validating the rules associated to these constraints. The approach to accomplish this is to spend more time in each node of the solutions graph by reducing the sets of possible values for not-yet-assigned variables. The model is introduced as a general control mechanism and realizes an a priori pruning in the solutions graph. This is done by assuming that the only rules to be considered are those arising from the propagation of their constraints and by computing only the rules that acquired some domain-dependent information about the significance of various domain interactions.